Towards gaze-independent c-VEP BCI: A pilot study

TL;DR

This pilot study makes the first step towards a gaze-independent speller based on the code-modulated visual evoked potential (c-VEP), showing the promising feasibility of utilizing the c-VEP protocol for gaze-independent BCIs that use covert spatial attention when both stimuli flash simultaneously.

Abstract

A limitation of brain-computer interface (BCI) spellers is that they require the user to be able to move the eyes to fixate on targets. This poses an issue for users who cannot voluntarily control their eye movements, for instance, people living with late-stage amyotrophic lateral sclerosis (ALS). This pilot study makes the first step towards a gaze-independent speller based on the code-modulated visual evoked potential (c-VEP). Participants were presented with two bi-laterally located stimuli, one of which was flashing, and were tasked to attend to one of these stimuli either by directly looking at the stimuli (overt condition) or by using spatial attention, eliminating the need for eye movement (covert condition). The attended stimuli were decoded from electroencephalography (EEG) and classification accuracies of 88% and 100% were obtained for the covert and overt conditions, respectively. These fundamental insights show the promising feasibility of utilizing the c-VEP protocol for gaze-independent BCIs that use covert spatial attention when both stimuli flash simultaneously.

Figures (3)

• Figure 1: Stimulus protocol. In (a), a graphical representation of the stimulus interface is depicted, featuring two stimuli positioned at $2.1^{\circ}$ on either side of a fixation cross. The stimuli took the form of circles measuring $3^{\circ}$ in both height and width. The fixation cross was $0.7^{\circ}$ for each side. The shapes presented were bound to a maximum height and width of $1.4^{\circ}$ each. The shapes' heights and widths were as follows: green circle ($1.4^{\circ}$ diameter), inverted cyan triangle and yellow triangle ($0.9^{\circ} \times 1.4^{\circ}$), magenta hourglass ($0.9^{\circ} \times 1.4^{\circ}$) and the red rectangle ($1.5^{\circ} \times 0.5^{\circ}$, rotated by $45^{\circ}$). In (b), a graphical representation of the stimulus protocol is depicted comprising two crucial components: first, the background of the stimuli underwent alternating black-and-white transitions following a binary pseudo-random sequence; second, diverse-colored shapes were presented within the stimuli. The stimulus background could dynamically change with each frame of 16.67 ms (60 Hz), while the shapes within the stimuli changed every 250 ms (4 Hz). A trial took 20 s, within which target shapes (the magenta hour glass) appeared randomly in the sequence with at least 1 s distance. Participants engaged with the stimuli by counting the number of target shapes on the attended side. In this pilot study, we adopted a paradigm where only the background of the attended stimulus alternated, while the background of the unattended stimulus remained constant. A left-attended trial is shown in (b).
• Figure 2: Classification accuracy across modeled transient response lengths. Depicted are the participant-specific classification accuracies for both overt (solid lines) and covert (dashed lines) conditions across transient response lengths ranging from 0.1 s to 0.9 s. The grand average over participants is shown in black. Please note, that for the overt condition, the classification accuracy was $100$ % for all transient response lengths and all participants. The dashed gray line indicates theoretical chance level ($50$ %).
• Figure 3: Spatial activity pattern and transient responses of participant S4. (a) and (b) show the spatial activity pattern and transient responses of S4 for the overt and covert conditions, respectively. For all participants, the spatial activity for the overt condition was more focally distributed as compared to the more lateralized distribution seen for the covert condition. The spatial pattern $\mathbf{a} \in \mathbb{R}^C$ was estimated as $\mathbf{a}=\mathbf{w}^\top{\bm\Sigma}$, where ${\bm\Sigma} \in \mathbb{R}^{C \times C}$ is the spatial covariance matrix.